Research On Infrared Image Processing And Recognition Of Typical Targets

This paper takes the problem of target recognition in anti-tank battle under battlefield environment as the research background,and focuses on the pattern recognition field of infrared image,develops the method of target recognition for typical tanks in battlefield.At present,a lot of research has been done on target recognition both at home and abroad,but it is generally based on the visible image for target recognition.The research on infrared image is not enough,and the computation of target recognition algorithm is too large to be integrated in embedded system.This paper chooses Xilinx Zynq SoC heterogeneous platform as the development platform,optimizes the algorithm,and achieves the construction of the embedded infrared image tank target recognition system.In the aspect of recognition algorithm research,this paper studies the target recognition method based on feature extraction and machine learning.In the feature description of the target,three features of local binary pattern feature(LBP),Hu invariant moment feature and histogram of oriented gradient(HOG)are analyzed and studied.After comparing the three features,aiming at the tank target in the infrared image to be identified in this paper,the shape and wheel of the target are represented based on the gradient intensity and gradient oriented.In the aspect of machine learning,we focus on the adaptive boosting algorithm(AdaBoost),construct logistic regression classifier as a weak classifier,get strong classifier through iterative training,and carry out experimental verification under the PC platform to analyze the feasibility of the algorithm.In the FPGA hardware optimization of the algorithm,this paper proposes a hardware optimization design of the algorithm for the feature extraction of HOG based on the calculation characteristics of the FPGA platform.The optimized algorithm improves the division operation which is difficult to be realized by FPGA into the shift operation,which can better adapt to the calculation characteristics of FPGA,exploit the parallel calculation advantages of FPGA,and improve the performance of the system.In the construction of embedded system,this paper adopts the method of Hardware/Software Co-Design,and uses Vivado HLS high-level synthesis tool to encapsulate the optimized hog feature extraction algorithm as IP core.In the vivado development environment,block design is carried out to build the hardware engineering of embedded system.In the software,PetaLinux tool is used to transplant the embedded Linux Operating System,QT and OpenCV libraries are transplanted through cross compilation,VDMA IP core driver is written to complete the construction of the embedded infrared image tank target recognition system,and experiments are carried out on the built target recognition system.According to the experimental results,the performance of the system is analyzed and evaluated.